Mapping Clinically Relevant Dose×LET Differences for Skin Sparing with Spot Delete In Head‑and‑Neck Proton Therapy
Abstract
Purpose
In proton therapy, conventional DVH comparisons may be insensitive to planning changes intended to reduce LET‑associated biological risk in normal tissues, despite clinically observed reductions in toxicity. Dose–LET volume histograms (DLVHs) provide a joint view of physical dose and LET, enabling localization of plan differences in Dose×LET space. The purpose of this study was to the identify Dose×LET region in skin DLVHs most associated with toxicity reduction using the Spot Delete skin‑sparing technique in Head-and-Neck (HN) proton therapy patients.
Methods
15 HN patients were retrospectively planned with Spot Delete (SD) in the 5mm skin rind and without (Ref). Skin DLVH difference maps were analyzed per bin as ΔVolume(Dose,LET)=VSD−VRef. Voxelwise paired inference used Wilcoxon signed‑rank testing with Benjamini–Hochberg false‑discovery‑rate (FDR) correction to generate q‑value maps, alongside effect‑size mapping using Cohen’s d. Because voxelwise significance maps (even after FDR) were too spatially extensive for clinical interpretation, an interpretable contiguous region was extracted using combined criteria emphasizing high LET, negative ΔV, strong significance, large effect size, and cluster contiguity.
Results
Three region‑finding strategies were evaluated to reduce “speckle” and yield a planning‑relevant summary (threshold‑intersection with contiguity filtering, ridge‑and‑ribbon, and permutation cluster‑mass). The selected region definition (LET ≥ 3.0 keV/µm, Dose(RBE=1.1) ≥ 10 Gy, q ≤ 0.05, mean ΔV < 0, Cohen’s d ≤ −0.8, retaining the largest 4‑connected component) identified a single dominant contiguous cluster localized to high LET at clinically relevant skin dose. Summed ΔV within this region was consistently negative across patients and remained strongly significant after FDR correction (Wilcoxon q≈2×10⁻⁴), consistent with systematic reduction of high‑LET skin volume in SD plans.
Conclusion
Dose×LET difference mapping with FDR‑controlled inference, effect‑size gating, and contiguity filtering isolates an interpretable high‑LET/high‑dose region capturing the dominant DLVH difference associated with Spot Delete skin sparing in HN, supporting LET‑aware skin planning objectives.